A Dynamic Scheduling and Pricing Method for On-Site Services Based on Multidimensional Data and Machine Learning
By using synchronous shadow diffusion and forward diffusion of the original chain mirror and the placeholder chain mirror, combined with price proof by contradiction, the problem of identifying the order disturbance of the target order insertion on the subsequent service chain and the accuracy of pricing in the on-site service platform is solved. This achieves scheduling stability and price closure, and reduces the risk of subsequent order crowding.
Patent Information
- Application Number
- CN202611279164.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-21
- Publication Date
- 2026-11-17
AI Technical Summary
Existing on-site service platforms struggle to promptly identify the disruptions to the subsequent service chain caused by inserting target orders in multi-order continuous fulfillment scenarios. The disconnect between scheduling calculations and pricing calculations results in a lack of closed-loop verification between order dispatch results and order prices in terms of fulfillment stability and transaction constraints.
By mirroring the original chain, mirroring the placeholder chain, synchronous shadow diffusion, tail-end reverse diffusion, and price reversal proof, the service chain insertion verification, performance disturbance identification, and acceptance price verification are completed. DCRNN diffusion convolutional recurrent units are used to perform synchronous shadow diffusion and forward diffusion, generate acceptance prices and perform price reversal proof, and freeze or release candidate service provider connection edges.
It achieves stable scheduling and accurate pricing, reduces the problem of subsequent order congestion, improves service matching accuracy and fulfillment stability, and ensures that price acceptability and scheduling feasibility are closed within the same verification link.
Smart Images

Figure CN123117634A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of Internet service platform technology, and in particular to a method for dynamic scheduling and pricing of on-site services based on multidimensional data and machine learning. Background Technology
[0002] With the development of local life service platforms and on-demand fulfillment services, orders for on-demand services such as housekeeping, repairs, beauty services, cleaning, and nursing are increasingly being received, allocated, and priced uniformly by online platforms. Existing on-demand service platforms mostly use methods such as fixed pricing, rule-based order assignment, and machine learning-based rating-based order assignment to process orders. Common processing criteria include service distance, service type, service provider rating, order history, service time period, and regional supply and demand status. Some platforms introduce predictive models to calculate estimated arrival time, probability of a transaction, and reference price, and then select service providers based on the rating results.
[0003] The above-mentioned processing methods still have shortcomings in scenarios involving continuous fulfillment of multiple orders. Existing solutions mostly match target orders as independent tasks, lacking a verification process that pre-embeds target orders into the subsequent service chains of candidate service providers. This makes it difficult to detect in a timely manner the order disturbances caused by inserting target orders to the tail-end tasks waiting to be fulfilled. Scheduling calculations and pricing calculations are also easily separated. The timely arrival of a service provider does not mean that the acceptance price is acceptable to the target order status, and the fact that the price meets user expectations does not mean that the candidate service chain will not experience subsequent fulfillment crowding. Existing DCRNN-type spatiotemporal prediction structures are mostly used for unidirectional prediction of path status and time status, lacking synchronous comparative diffusion between the original chain mirror and the placeholder chain mirror. It is difficult to read the chain tail disturbance through tail-end back diffusion and then incorporate the chain tail disturbance into the acceptance price to form a price counter-proof. This results in a lack of closed-loop verification between the order dispatch results and the order price in terms of fulfillment stability, impact on subsequent tasks, and transaction constraints.
[0004] Therefore, how to provide dynamic scheduling and pricing methods for on-site services based on multidimensional data and machine learning is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0005] One objective of this invention is to propose a dynamic scheduling and pricing method for on-site services based on multidimensional data and machine learning. This invention completes service chain insertion verification, performance disturbance identification, and acceptance price verification through original chain mirroring, placeholder chain mirroring, synchronous shadow diffusion, tail-end reverse diffusion, and price inversion. It has the advantages of stable scheduling, accurate pricing, less subsequent order crowding, and high service matching accuracy.
[0006] The on-site service dynamic scheduling and pricing method based on multidimensional data and machine learning according to embodiments of the present invention includes the following steps: Obtain the target order status and candidate service chain, and construct a dynamic supply and demand graph containing candidate service provider connection edges according to the on-site service fulfillment order; The target order is embedded into the candidate service chain as a misaligned placeholder, and the service chain before embedding is mirrored as the original chain mirror. The original chain image and the placeholder chain image are fed into the DCRNN diffusing convolution recurrent unit. The original chain image is driven to perform synchronous shadow diffusion with the same diffusion parameters, and the placeholder chain image is driven to perform forward diffusion, so as to obtain the synchronous shadow diffusion result and the forward diffusion result. After forward diffusion reaches the misaligned occupant point, it triggers tail-end reverse diffusion, which returns from the tail end of the occupant chain mirror to the misaligned occupant point, and reads the chain tail perturbation formed by the occupant chain mirror relative to the original chain mirror at the chain tail node. After the chain tail disturbance is folded in, the acceptance price is generated. The acceptance price is then written back to the target order status through the misaligned occupancy point. When the acceptance price does not fall into the price acceptance range corresponding to the target order status, a price counter-proof is generated. For candidate service provider connection edges where the chain tail disturbance has not returned to zero, the performance freeze will be executed; for candidate service provider connection edges where there is a price counter-proof, the pricing freeze will be executed; for candidate service provider connection edges where the chain tail disturbance has returned to zero and there is no price counter-proof, the connection will be released. After the connection is released, perform residual deduction between the synchronous shadow diffusion result and the forward diffusion result on the candidate service provider connection edge, and generate the service provider allocation result and order price according to the residual deduction result.
[0007] Optionally, the process of constructing the dynamic supply and demand graph specifically includes: Using the target order status as the node to be inserted, candidate service chains are developed according to the order of execution of unfulfilled tasks; Write the unfinished tasks in the candidate service chain as chain nodes in sequence, and connect adjacent chain nodes according to the order of execution. Mark insertable chain segments between adjacent chain nodes along the candidate service chain, and preserve the order of performance between adjacent chain nodes; The node to be inserted is attached to the front and back chain nodes of the insertable chain segment, respectively, so that the insertable chain segment forms a channel to be verified. Bind the channel to be verified to the candidate service provider's connection edge, and write the verification status to the candidate service provider's connection edge; Candidate service chains that have not yet formed a verification channel are prohibited from entering the mirroring process of the original chain mirror and the placeholder chain mirror.
[0008] Optionally, embedding the target order as a misaligned occupant point into the candidate service chain specifically includes: Read the candidate service provider connection edge that is in the pending verification state and locate the pending verification channel bound to the candidate service provider connection edge; Read the front-end chain node and the back-end chain node along the channel to be verified, and push the node to be inserted between the front-end chain node and the back-end chain node; Mark the inserted node as a misalignment occupant point and maintain the order of fulfillment from the front-end chain node to the back-end chain node without deleting it; Write a front-end placeholder sequential connection between the front-end chain node and the misaligned placeholder point, and write a back-end placeholder sequential connection between the misaligned placeholder point and the back-end chain node. Process the unverified channels in the candidate service chain one by one according to the order of performance, so that each unverified channel forms a placeholder candidate chain, and bind the placeholder candidate chain to the corresponding candidate service provider connection edge.
[0009] Optionally, the step of mirroring the pre-embedded service chain as the original chain image and mirroring the post-embedded service chain as a placeholder chain image specifically includes: Read the candidate service chain and placeholder candidate chain bound to the candidate service provider connection edge; Copy all chain nodes and fulfillment order connections in the candidate service chain to obtain a mirror image of the original chain; Copy all chain nodes, misaligned placeholder points, fulfillment order connections, front-end placeholder order connections, and back-end placeholder order connections in the placeholder candidate chain to obtain a placeholder chain mirror image. Bind the original chain image and the placeholder chain image to the same candidate service provider connection edge; Write the same chain node sequence number to the original chain mirror and the placeholder chain mirror according to the order of performance; Write an empty mark at the corresponding misaligned position of the chain node in the original chain mirror, and retain the order of the chain nodes before and after the empty mark so that the original chain mirror and the occupied chain mirror form a same-order comparison structure. The same-order comparison structure is bound to the candidate service provider connection edge and written into the same-parameter input queue of the DCRNN diffusing convolution recurrent unit.
[0010] Optionally, obtaining the synchronous shadow diffusion result and the forward diffusion result specifically includes: Read the same-order comparison structure of the connection edges bound to the same candidate service provider in the same parameter input queue; Lock the diffusion rounds, diffusion step size, diffusion order, and diffusion weight of the original chain mirror and the placeholder chain mirror; The original chain mirror is driven to perform synchronous shadow diffusion along the fulfillment order connection. When diffusion reaches the empty slot mark, the fulfillment order connection of the chain nodes before and after the empty slot mark is maintained. The drive placeholder chain mirrors are connected along the fulfillment sequence, the front placeholder sequence is connected, the staggered placeholder points are connected, and the rear placeholder sequence is connected to perform forward diffusion; At the end of each diffusion round, the original chain mirror diffusion state and the placeholder chain mirror diffusion state under the same chain node number are synchronously latched. Diffusion states with different chain node numbers are prohibited from being written to the same latch location. Write the original chain mirror diffusion state after synchronization latch as the synchronization shadow diffusion result, and write the placeholder chain mirror diffusion state after synchronization latch as the forward diffusion result.
[0011] Optionally, the generation process of the chain tail perturbation specifically includes: Capture the diffusion round in which the forward diffusion result first hits the misaligned site, and lock that diffusion round as the tail-end reverse diffusion trigger round; At the tail end, the reverse diffusion triggers a round of latching the arrival position corresponding to the forward diffusion result, and prohibits the forward diffusion result after the misaligned occupancy point from being directly written into the chain tail perturbation; Using the tail node of the placeholder chain mirror as the reverse diffusion entry point, connect along the fulfillment sequence of the placeholder chain mirror and then connect the subsequent placeholder sequence back to the misaligned placeholder point. At the tail node, the forward diffusion result is sequentially merged with the synchronous shadow diffusion result under the same chain node number; Write the performance sequence offset, performance start offset, and performance completion offset formed by the same sequence of snapping into the chain tail disturbance; When the fulfillment sequence offset, fulfillment start offset, and fulfillment completion offset are all returned to zero, the chain tail disturbance is written as a zero state; If any of the following offsets—performance sequence offset, performance start offset, or performance completion offset—is not zeroed, the chain tail disturbance will be written as not zeroed.
[0012] Optionally, the generation of price disproven proof specifically includes: Read one of the zeroed-out or non-zeroed-out states of the chain tail perturbation; When the chain tail disturbance is in a zero state, the basic acceptance price corresponding to the target order status is written into the acceptance price; When the chain tail disturbance is not in a zero state, write the basic acceptance price corresponding to the target order status item by item in the order of performance sequence offset, performance start offset, and performance completion offset to obtain the acceptance price; The accepted price is written back to the target order status via the misaligned occupant point, locking in the price acceptance range corresponding to the target order status; Perform a closed-loop comparison between the accepted price and the price range; When the price falls into the price acceptance range, write the price reversal to the empty state; When the accepting price is lower than the lower boundary of the price acceptance range, write the low-price overshoot status into the price proof; when the accepting price exceeds the upper boundary of the price acceptance range, write the high-price overshoot status into the price proof.
[0013] Optionally, the freeze-release process of the candidate service provider connection edge specifically includes: Read the tail perturbation and price reversal evidence bound to the candidate service provider's connection edge; When the chain tail disturbance is not in a zero state, the candidate service provider connection edge is written to the performance freeze state to prevent the candidate service provider connection edge from entering the residual deduction. When the price counter-proof is written into either the low price out of bounds state or the high price out of bounds state, the candidate service provider connection edge is written into the pricing freeze state to prevent the candidate service provider connection edge from entering the residual deduction. When the chain tail disturbance is in a zero state and the price proof is empty, clear the unverified state of the candidate service provider connection edge and write the candidate service provider connection edge into the connection release state. When a candidate service provider's connection edge is written to either the performance freeze state or the pricing freeze state, it remains frozen and connection release is prohibited. Only candidate service provider connection edges in the connection release state are allowed to enter the residual subtraction of synchronous shadow diffusion results and forward diffusion results.
[0014] Optionally, the step of generating service provider allocation results and order prices based on residual deduction results specifically includes: Using candidate service provider connection edges in the connection release state as the snapping objects, extract the synchronous shadow diffusion results and forward diffusion results bound to the same candidate service provider connection edges; Press the synchronous shadow diffusion results and forward diffusion results into the same sequence slot according to the same chain node number, and prohibit the out-of-order snapping between different chain node numbers; Using the vacancy mark corresponding to the chain node number as the residual entry, the vacancy propagation state of the misplaced vacancy point in the forward diffusion result is matched with the shadow propagation state of the chain nodes before and after the vacancy mark in the synchronous shadow diffusion result. Write the propagation offset that is not fully locked after locking into the residual locking result. When the propagation offset returns to zero, write the residual locking result into the residual closed state. Perform edge selection processing on the candidate service provider connection edges that are written to the residual closure state, and write the candidate service providers corresponding to the selected candidate service provider connection edges into the service provider allocation results; The acceptance price corresponding to the selected candidate service provider's connection edge is written into the order price.
[0015] Optionally, the edge selection process specifically includes: Read and write candidate service provider connection edges in the residual closure state; Candidate service provider connection edges are written into the closed sequence according to the order in which the residual closure state is written, and candidate service provider connection edges that have not been written into the residual closure state are prohibited from entering the closed sequence. When there are several candidate service provider connection edges in the same writing order, the candidate service provider connection edge whose residual entry is written first is selected according to the writing order of the residual entry in the same sequence locking slot. The selected candidate service providers will be locked as the target connection edges; Write the candidate service providers corresponding to the target connection edge into the service provider allocation results; Write the acceptance price corresponding to the target connection edge into the order price.
[0016] The beneficial effects of this invention are: (1) In this invention, the target order is embedded into the candidate service chain as a misaligned placeholder point, and the original chain image and the placeholder chain image are retained. Synchronous shadow diffusion and forward diffusion are performed under the same diffusion parameters, and the chain tail disturbance is read by the reverse folding back from the tail end of the placeholder chain image. Through this process, the platform can identify the sequence offset, start offset and completion offset caused by the insertion of the target order to the subsequent fulfillment tasks before dispatching the order, thus avoiding the problem of subsequent order crowding caused by selecting service providers based solely on distance, score or expected arrival time.
[0017] (2) The present invention incorporates the chain tail disturbance into the acceptance price and writes the acceptance price back to the target order status through the misaligned occupancy point, and compares it with the price acceptance range to generate a price counter-proof. Through this process, the order price is no longer determined solely by the base price, distance cost, or supply and demand changes, but simultaneously absorbs the performance disturbance generated after the target order is inserted into the candidate service chain, so that scheduling feasibility and price acceptability are closed within the same verification link, reducing the situation of "orders can be dispatched but prices are not valid" or "prices are valid but performance is squeezed out".
[0018] (3) The present invention performs performance freezing on candidate service provider connection edges whose chain tail disturbance has not returned to zero, and performs pricing freezing on candidate service provider connection edges with price counter-proof. Only candidate service provider connection edges in the connection release state are allowed to enter the residual deduction of synchronous shadow diffusion results and forward diffusion results. Through this processing, the service provider allocation results and order prices are derived from the residual closure state after mirror diffusion, rather than a single score ranking, which improves the stability of service chain insertion, pricing closure and consistency of final scheduling results. Attached Figure Description
[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the on-site service dynamic scheduling and pricing method based on multidimensional data and machine learning proposed in this invention; Figure 2 This is a schematic diagram of the misaligned occupancy and double-mirror isoparametric diffusion proposed in this invention; Figure 3 This is a schematic diagram of the tail-end reverse diffusion, price reversal proof, and residual edge selection proposed in this invention. Detailed Implementation
[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.
[0021] refer to Figures 1-3 The method for dynamic scheduling and pricing of on-site services based on multidimensional data and machine learning includes the following steps: Obtain the target order status and candidate service chain, and construct a dynamic supply and demand graph containing candidate service provider connection edges according to the on-site service fulfillment order; In this invention, the target order status is the order-side status record formed when the target order enters the scheduling and pricing process, which is generated by the platform after receiving the on-site service request; the target order status includes service type, service location, appointment time, service duration requirement, and price acceptance range, and is used as a node to be inserted in the dynamic supply and demand diagram; The candidate service chain is a chain structure formed by arranging the unfulfilled tasks of candidate service providers in the order of their on-site service performance. The candidate service chain is generated from the task records that the candidate service providers have undertaken but have not completed. Each unfulfilled task corresponds to a chain node, and adjacent chain nodes are connected by the performance order to form a chain arrangement. The order of on-site service fulfillment is the order in which candidate service providers execute unfulfilled tasks; the order of on-site service fulfillment is arranged according to the confirmed appointment time; if the confirmed appointment times are the same, they are arranged according to the platform's order dispatch time; if the platform's order dispatch time is the same, they are arranged according to the time the task is written into the candidate service chain. The candidate service provider connection edge is a connection to be verified between the target order status and the candidate service chain. The set of states is fixed as follows: pending verification, fulfillment freeze, pricing freeze, connection release, and residual closure. The pending verification state indicates that the connection edge has not yet completed the chain tail perturbation and price reversal verification. The fulfillment freeze state indicates that the chain tail perturbation has not been returned to zero. The pricing freeze state indicates that the price reversal has been written into one of the low price out-of-bounds state or the high price out-of-bounds state. The connection release state indicates that the chain tail perturbation is in the zero state and the price reversal is empty. The residual closure state indicates that the propagation offset in the residual deduction result is returned to zero. The dynamic supply and demand graph is a directed graph structure that carries the target order status, candidate service chains, and candidate service provider connection edges. The dynamic supply and demand graph consists of nodes to be inserted, chain nodes, fulfillment order connections, and candidate service provider connection edges, and is used to carry subsequent misalignment and placement, mirroring processing, DCRNN diffusion, and connection edge freezing and release. The target order is embedded into the candidate service chain as a misaligned placeholder, and the service chain before embedding is mirrored as the original chain mirror. In this invention, the misaligned occupant point is the temporary insertion position of the target order in the candidate service chain, which is obtained by marking the node to be inserted between the front-end chain node and the back-end chain node; the misaligned occupant point is used to record the occupant position formed after the target order is inserted into the candidate service chain; the front-end chain node is the chain node before the insertion position, and the back-end chain node is the chain node after the insertion position. The pre-embedded service chain is a chain structure when the target order is not written into the candidate service chain, retaining the chain nodes and fulfillment order connections in the candidate service chain; the post-embedded service chain is a chain structure formed after the target order is written into the candidate service chain as a misaligned placeholder, including misaligned placeholders, front-end placeholder order connections, and back-end placeholder order connections.
[0022] The original chain mirror is a copy of the service chain before embedding, and is obtained by copying the service chain before embedding; the original chain mirror is used to retain the fulfillment propagation reference when the target order is not inserted; the placeholder chain mirror is a copy of the service chain after embedding, and is obtained by copying the service chain after embedding; the placeholder chain mirror is used to represent the fulfillment propagation path after the target order is inserted. The mirroring process replicates the chain nodes and connections of the service chain, and establishes a comparison operation based on the same chain node sequence number. After mirroring, the empty slots in the original chain mirror and the misaligned slots in the placeholder chain mirror are located at the same chain node sequence number. The original chain image and the placeholder chain image are fed into the DCRNN diffusing convolution recurrent unit. The original chain image is driven to perform synchronous shadow diffusion with the same diffusion parameters, and the placeholder chain image is driven to perform forward diffusion, so as to obtain the synchronous shadow diffusion result and the forward diffusion result. In this invention, the DCRNN diffusing convolutional recurrent unit is a graph sequence processing unit in the diffusing convolutional recurrent neural network, used to propagate the state of chain nodes along the connection relationship of candidate service chains; the state of a chain node is a record of the service performance status of the chain node in the diffusion round, including the propagation information corresponding to the arrival order, start time and completion time; The same diffusion parameters are the diffusion round, diffusion step size, diffusion order, and diffusion weight shared by the original chain mirror and the placeholder chain mirror; the original chain mirror and the placeholder chain mirror use the same diffusion parameters to enter the DCRNN diffusion convolution recurrent unit, so that the two mirror chains form comparable diffusion results under the same chain node index. Synchronous shadow diffusion is a reference propagation operation on the original chain mirror. During execution, the original chain mirror connects the propagation chain node states along the fulfillment order. When an empty slot is encountered, the target order state is not written. The fulfillment order connection of the chain nodes before and after the empty slot is retained. After the diffusion is completed, the synchronous shadow diffusion result is obtained. Forward diffusion is a placeholder propagation operation on the placeholder chain mirror; during execution, the placeholder chain mirror is connected along the fulfillment order, the front placeholder order is connected, through the misaligned placeholder point, and then along the rear placeholder order to connect the propagation chain node states. After the diffusion is completed, the forward diffusion result is obtained. The synchronous shadow diffusion result is the diffusion state record formed by the original chain mirror in the diffusion round, and the forward diffusion result is the diffusion state record formed by the placeholder chain mirror in the same diffusion round; the two types of diffusion results correspond to the same chain node number and are used for subsequent tail-end reverse diffusion, chain tail perturbation reading and residual deduction. After forward diffusion reaches the misaligned occupant point, it triggers tail-end reverse diffusion, which returns from the tail end of the occupant chain mirror to the misaligned occupant point, and reads the chain tail perturbation formed by the occupant chain mirror relative to the original chain mirror at the chain tail node. In this invention, the forward diffusion reaching the misaligned occupant site is the trigger state formed after the forward diffusion result is written into the corresponding chain node number of the misaligned occupant site; when the DCRNN diffusion convolution recurrent unit propagates along the occupant chain in a mirror image, it records the diffusion round and arrival position of the forward diffusion result hitting the misaligned occupant site, and uses the hit time as the trigger basis for the tail-end reverse diffusion. Tail-end reverse diffusion is a diffusion verification operation that reverses from the tail node of the placeholder chain mirror to the misplaced placeholder point. During execution, the last chain tail node in the fulfillment order of the placeholder chain mirror is used as the reverse diffusion entry point. The chain node sequence number is read in reverse along the subsequent placeholder order connection and fulfillment order connection until it returns to the misplaced placeholder point. After the chain tail disturbance is folded in, the acceptance price is generated. The acceptance price is then written back to the target order status through the misaligned occupancy point. When the acceptance price does not fall into the price acceptance range corresponding to the target order status, a price counter-proof is generated. In this invention, the acceptance price is the price formed after the target order is verified by the candidate service provider's connection edge; the acceptance price is obtained by combining the base acceptance price corresponding to the target order status and the chain tail perturbation. The chain tail disturbance is incorporated into the processing operation of writing the chain tail disturbance into the base acceptance price; when the chain tail disturbance is in a zero state, the base acceptance price is written into the acceptance price; when the chain tail disturbance is not in a zero state, the performance order offset, performance start offset, and performance completion offset are written into the base acceptance price to obtain the acceptance price. The acceptance price is written back to the target order status via the misaligned placeholder point. In order to return the acceptance price to the target order status along the misaligned placeholder point in the placeholder chain mirror, the target order status is compared with the price acceptance range based on the inserted candidate service chain. The price acceptance range is the acceptable price range recorded in the target order status, consisting of the lower boundary and the upper boundary of the price acceptance range. When the accepting price is between the lower boundary and the upper boundary of the price acceptance range, the price rebuttal is written to the empty state. When the accepting price is lower than the lower boundary of the price acceptance range, the price rebuttal is written to the low price out of bounds state. When the accepting price exceeds the upper boundary of the price acceptance range, the price rebuttal is written to the high price out of bounds state. Price reversal is the pricing verification result formed when the acceptance price does not meet the price acceptance range corresponding to the target order status, and is used for subsequent pricing freeze judgment of candidate service provider connection edges; For candidate service provider connection edges where the chain tail disturbance has not returned to zero, the performance freeze will be executed; for candidate service provider connection edges where there is a price counter-proof, the pricing freeze will be executed; for candidate service provider connection edges where the chain tail disturbance has returned to zero and there is no price counter-proof, the connection will be released. In this invention, the performance freeze is a restriction state written to the candidate service provider connection edge when the chain tail disturbance is not zeroed; when the chain tail disturbance is not zeroed, it means that after the target order is inserted into the candidate service chain, at least one of the following occurs to the chain tail node: performance order offset, performance start offset, and performance completion offset. The platform writes the corresponding candidate service provider connection edge into the performance freeze state to prevent the candidate service provider connection edge from entering the residual deduction. Price freeze is a restriction state written to the candidate service provider's connection edge when price counter-proof exists; price counter-proof exists when price counter-proof is written to either the low price out of bounds state or the high price out of bounds state; when the above states are met, the platform will write the corresponding candidate service provider's connection edge to the price freeze state, preventing the candidate service provider's connection edge from entering residual deduction. The "Connection Release" state is the state that a candidate service provider's connection edge can participate in when the chain tail perturbation is zero and the price is falsely proven to be empty. When the chain tail perturbation is zero and the price is falsely proven to be empty, it means that after the target order is inserted into the candidate service chain, it does not cause a fulfillment order offset, fulfillment start offset, or fulfillment completion offset to the chain tail node, and the acceptance price is within the price acceptance range. The platform clears the pending verification state of the candidate service provider's connection edge and writes the candidate service provider's connection edge into the "Connection Release" state. The candidate service provider's connection edge in the "Connection Release" state enters the residual deduction between the synchronous shadow diffusion result and the forward diffusion result. After the connection is released, perform residual deduction between the synchronous shadow diffusion result and the forward diffusion result on the candidate service provider connection edge, and generate the service provider allocation result and order price according to the residual deduction result.
[0023] In this invention, a candidate service provider connection edge after connection release is a candidate service provider connection edge whose tail disturbance is in a zero state and whose price is in a negative proof state. The connection release state indicates that the candidate service provider connection edge is qualified to enter residual deduction.
[0024] Residual decoupling is a decoupling operation performed between the synchronous shadow diffusion result and the forward diffusion result at the same chain node index. During execution, the chain node index corresponding to the empty position mark in the original chain mirror is used as the decoupling position. The forward diffusion result corresponding to the misaligned position in the placeholder chain mirror is decoupled with the synchronous shadow diffusion result corresponding to the chain nodes before and after the empty position mark in the original chain mirror. The propagation offset of any incomplete decoupling is written into the residual decoupling result. The propagation offset represents the difference in state between the forward diffusion result and the synchronous shadow diffusion result at the same chain node index when decoupling is incomplete.
[0025] When the propagation offset in the residual deduction result is zero, it means that the insertion of the target order into the candidate service chain has not disrupted the same-order correspondence between the original chain mirror and the placeholder chain mirror. The corresponding candidate service provider connection edge is used to generate the service provider allocation result and the order price.
[0026] The service provider allocation result consists of candidate service provider records determined based on the residual deduction result. The order price is the acceptance price bound to the connection edge of the candidate service provider corresponding to the residual deduction result. During generation, the candidate service providers bound to the connection edge of the candidate service provider corresponding to the residual deduction result are written into the service provider allocation result, and the acceptance price bound to the connection edge of the candidate service provider corresponding to the residual deduction result is written into the order price.
[0027] In this embodiment, the process of constructing the dynamic supply and demand map specifically includes: Using the target order status as the node to be inserted, candidate service chains are developed according to the order of execution of unfulfilled tasks; In this invention, an unfulfilled task refers to an on-site service task that a candidate service provider has accepted but has not yet completed. Unfulfilled tasks originate from the candidate service provider's task records. When a task is in one of the following states: accepted, booked, or in service, it is written into the candidate service chain. When a task is in one of the following states: completed, cancelled, or refunded, it is not written into the candidate service chain. The order of execution of the task is consistent with the order of on-site service fulfillment, both referring to the order in which candidate service providers execute unfulfilled tasks. Write the unfinished tasks in the candidate service chain as chain nodes in sequence, and connect adjacent chain nodes according to the order of execution. Mark insertable chain segments between adjacent chain nodes along the candidate service chain, and preserve the order of performance between adjacent chain nodes; In this invention, an insertable chain segment is a chain segment in a candidate service chain that allows the insertion of a node between adjacent chain nodes; the insertable chain segment is determined by the time order of the front-end chain node, the back-end chain node, and the node to be inserted; when the service completion time of the front-end chain node is not later than the reservation time of the node to be inserted, and the reservation time plus service duration of the node to be inserted is not later than the service start time of the back-end chain node, adjacent chain nodes are marked as insertable chain segments; the front-end chain node is the chain node preceding the adjacent chain node in terms of performance execution, and the back-end chain node is the chain node following the adjacent chain node in terms of performance execution. The node to be inserted is attached to the front and back chain nodes of the insertable chain segment, respectively, so that the insertable chain segment forms a channel to be verified. In this invention, the verification channel is a temporary verification path formed after the node to be inserted is attached to the insertable chain segment; the verification channel is used to carry the insertion verification between the target order status and the candidate service chain, without changing the original fulfillment order connection between adjacent chain nodes; Bind the channel to be verified to the candidate service provider's connection edge, and write the verification status to the candidate service provider's connection edge; In this invention, the state to be verified is the initial state of the candidate service provider connection edge before it enters the chain tail perturbation and price reversal verification; after the candidate service provider connection edge is written into the state to be verified, it enters the misalignment, mirroring and DCRNN diffusion convolution recurrent unit processing; the candidate service provider connection edge that is not written into the state to be verified does not enter the misalignment, mirroring and DCRNN diffusion convolution recurrent unit processing. Candidate service chains that have not yet formed a verification channel are prohibited from entering the mirroring process of the original chain mirror and the placeholder chain mirror.
[0028] In this invention, candidate service chains that have not formed a verification channel are prohibited from entering the mirroring process; this is a screening operation before mirroring. When a candidate service chain has not formed a verification channel, it means that the target order status cannot be attached to the insertable segment of the candidate service chain, and the original chain mirror and the placeholder chain mirror are no longer generated.
[0029] In this embodiment, the target order is embedded as a misaligned occupant point in the candidate service chain, specifically including: Read the candidate service provider connection edge that is in the pending verification state and locate the pending verification channel bound to the candidate service provider connection edge; In this invention, the candidate service provider's connection edge is bound to the channel to be verified. In order to read the binding relationship between the candidate service provider's connection edge and the channel to be verified, the front-end chain node and back-end chain node to be inserted into the target order are determined. The front-end chain node and the back-end chain node come from the adjacent chain nodes at both ends of the channel to be verified. Read the front-end chain node and the back-end chain node along the channel to be verified, and push the node to be inserted between the front-end chain node and the back-end chain node; In this invention, pushing is a temporary insertion operation that places the node to be inserted between the front-end chain node and the back-end chain node; after pushing, the node to be inserted is marked as a misalignment occupant point, and the original performance order connection from the front-end chain node to the back-end chain node is kept and not deleted, which is used to preserve the original performance order connection of the candidate service chain. Mark the inserted node as a misalignment occupant point and maintain the order of fulfillment from the front-end chain node to the back-end chain node without deleting it; Write a front-end placeholder sequential connection between the front-end chain node and the misaligned placeholder point, and write a back-end placeholder sequential connection between the misaligned placeholder point and the back-end chain node. In this invention, the front-end placeholder sequence connection is a placeholder connection written between the front-end chain node and the misaligned placeholder point, and the back-end placeholder sequence connection is a placeholder connection written between the misaligned placeholder point and the back-end chain node; the front-end placeholder sequence connection and the back-end placeholder sequence connection are used to represent the connection relationship between the target order and the candidate service chain after the target order is inserted. Process the unverified channels in the candidate service chain one by one according to the order of performance, so that each unverified channel forms a placeholder candidate chain, and bind the placeholder candidate chain to the corresponding candidate service provider connection edge.
[0030] In this invention, the placeholder candidate chain is a chain structure formed by inserting staggered placeholder points into the channel to be verified. The placeholder candidate chain includes the original chain nodes in the candidate service chain, staggered placeholder points, fulfillment order connections, front-end placeholder order connections, and back-end placeholder order connections. Each channel to be verified forms a separate placeholder candidate chain and is bound to the corresponding candidate service provider connection edge.
[0031] In this embodiment, the service chain before embedding is mirrored as the original chain image, and the service chain after embedding is mirrored as a placeholder chain image, specifically including: Read the candidate service chain and placeholder candidate chain bound to the candidate service provider connection edge; Copy all chain nodes and fulfillment order connections in the candidate service chain to obtain a mirror image of the original chain; In this invention, replication is a mirror generation operation that preserves the chain nodes, connection relationships, and execution order; after replicating the candidate service chain, the original chain mirror is obtained, and after replicating the placeholder candidate chain, the placeholder chain mirror is obtained; the replication process does not change the original connection relationships in the candidate service chain and the placeholder candidate chain. Copy all chain nodes, misaligned placeholder points, fulfillment order connections, front-end placeholder order connections, and back-end placeholder order connections in the placeholder candidate chain to obtain a placeholder chain mirror image. Bind the original chain image and the placeholder chain image to the same candidate service provider connection edge; Write the same chain node sequence number to the original chain mirror and the placeholder chain mirror according to the order of performance; In this invention, the chain node sequence number is the sequential number written into the chain nodes according to the order of performance; the original chain mirror and the placeholder chain mirror use the same chain node sequence number, so that the position corresponding to the same chain node sequence number is consistent in the two mirror chains. Write an empty mark at the corresponding misaligned position of the chain node in the original chain mirror, and retain the order of the chain nodes before and after the empty mark so that the original chain mirror and the occupied chain mirror form a same-order comparison structure. In this invention, the empty slot marker is the position marker of the corresponding misplaced occupant chain node in the original chain mirror; the empty slot marker indicates that the target order has not been inserted into the original chain mirror, and the chain nodes before and after the empty slot marker are still connected in the original fulfillment order; The same-order comparison structure is a comparison structure formed by the original chain mirror and the placeholder chain mirror according to the same chain node number; the same-order comparison structure is used to make the empty mark and the misplaced placeholder point located at the same chain node number, so that the subsequent synchronous shadow diffusion results and forward diffusion results can be compared according to the same chain node number. The same-order comparison structure is bound to the candidate service provider connection edge and written into the same-parameter input queue of the DCRNN diffusing convolution recurrent unit.
[0032] In this invention, the isoparametric input queue is the input queue for the DCRNN diffusing convolutional recurrent unit to receive the same sequence of reference structures. In the isoparametric input queue, the original chain mirror and the placeholder chain mirror bound to the same candidate service provider connection edge use the same diffusing round, diffusing step size, diffusing order, and diffusing weight.
[0033] In this embodiment, the synchronous shadow diffusion result and the forward diffusion result are obtained, specifically including: Read the same-order comparison structure of the connection edges bound to the same candidate service provider in the same parameter input queue; Lock the diffusion rounds, diffusion step size, diffusion order, and diffusion weight of the original chain mirror and the placeholder chain mirror; In this invention, the diffusion rounds, diffusion step size, diffusion order, and diffusion weights of the original chain mirror and the placeholder chain mirror are locked to ensure that the same diffusion control conditions are applied to the two mirror chains within the DCRNN diffusion convolution recurrent unit. The diffusion round is the number of times the chain node state is updated, the diffusion step size is the number of connection steps traversed in each state propagation, the diffusion order is the number of connection layers allowed to be traversed along the connection relationship in each diffusion round, and the diffusion weight is the weight of the fulfillment order connection, the front-end placeholder order connection, and the back-end placeholder order connection participating in state propagation. The diffusion weights are written by the same parameter input queue, and the original chain mirror and the placeholder chain mirror bound to the same candidate service provider connection edge use the same diffusion weights. The original chain mirror is driven to perform synchronous shadow diffusion along the fulfillment order connection. When diffusion reaches the empty slot mark, the fulfillment order connection of the chain nodes before and after the empty slot mark is maintained. The drive placeholder chain mirrors are connected along the fulfillment sequence, the front placeholder sequence is connected, the staggered placeholder points are connected, and the rear placeholder sequence is connected to perform forward diffusion; At the end of each diffusion round, the original chain mirror diffusion state and the placeholder chain mirror diffusion state under the same chain node number are synchronously latched. Diffusion states with different chain node numbers are prohibited from being written to the same latch location. In this invention, the original chain mirror diffusion state is the chain node state record formed after each diffusion round of the original chain mirror; the placeholder chain mirror diffusion state is the chain node state record formed after each diffusion round of the placeholder chain mirror; the chain node state record contains propagation information corresponding to the chain node sequence number, arrival order, start time, and completion time. Synchronous latching is a processing operation that writes the original chain mirror diffusion state and the placeholder chain mirror diffusion state into the corresponding latch position according to the same chain node number at the end of the same diffusion round. The latch position is the state storage position set according to the chain node number in the DCRNN diffusion convolution loop unit. One chain node number corresponds to one latch position, and the diffusion state of different chain node numbers must not be written into the same latch position. Write the original chain mirror diffusion state after synchronization latch as the synchronization shadow diffusion result, and write the placeholder chain mirror diffusion state after synchronization latch as the forward diffusion result.
[0034] In this invention, the synchronous shadow diffusion result is the original chain's mirror diffusion state after synchronous latching, and the forward diffusion result is the placeholder chain's mirror diffusion state after synchronous latching. The synchronous shadow diffusion result and the forward diffusion result correspond to the same chain node number and are used for subsequent tail-end reverse diffusion, chain tail perturbation generation, and residual deduction.
[0035] In this embodiment, the process of generating the chain tail perturbation specifically includes: Capture the diffusion round in which the forward diffusion result first hits the misaligned site, and lock that diffusion round as the tail-end reverse diffusion trigger round; In this invention, the first hit on the misaligned occupant site is the state where the forward diffusion result is first written to the chain node number corresponding to the misaligned occupant site; the tail-end reverse diffusion trigger round is the diffusion round corresponding to the first hit on the misaligned occupant site by the forward diffusion result, which is used to start the reverse back-turn processing from the chain tail node to the misaligned occupant site. At the tail end, the reverse diffusion triggers a round of latching the arrival position corresponding to the forward diffusion result, and prohibits the forward diffusion result after the misaligned occupancy point from being directly written into the chain tail perturbation; In this invention, the arrival position is the chain node number corresponding to when the forward diffusion result hits the misaligned occupant site, and the position of the misaligned occupant site in the occupant chain mirror; the forward diffusion result after the misaligned occupant site is prohibited from being directly written into the chain tail perturbation, so that the chain tail perturbation only comes from the same sequence snapping result at the chain tail node. Using the tail node of the placeholder chain mirror as the reverse diffusion entry point, connect along the fulfillment sequence of the placeholder chain mirror and then connect the subsequent placeholder sequence back to the misaligned placeholder point. The reverse diffusion entry point is the position where the tail-end reverse diffusion begins reading; in this invention, the reverse diffusion entry point is fixed as the tail node in the placeholder chain mirror. The tail-end reverse diffusion reads the chain node number in reverse order along the fulfillment sequence connection and the subsequent placeholder sequence connection until it returns to the misaligned placeholder point. At the tail node, the forward diffusion result is sequentially merged with the synchronous shadow diffusion result under the same chain node number; In this invention, sequential fastening is a process of fastening the forward diffusion result at the tail node of the chain with the synchronous shadow diffusion result under the same chain node number; after sequential fastening, the fulfillment order offset, the fulfillment start offset and the fulfillment completion offset are formed and written into the tail disturbance. Write the performance sequence offset, performance start offset, and performance completion offset formed by the same sequence of snapping into the chain tail disturbance; In this invention, the fulfillment order offset is the difference in fulfillment sequence number formed by the tail node in the placeholder chain mirror relative to the original chain mirror; the fulfillment start offset is the difference in service start time formed by the tail node in the placeholder chain mirror relative to the original chain mirror; and the fulfillment completion offset is the difference in service completion time formed by the tail node in the placeholder chain mirror relative to the original chain mirror. When the fulfillment sequence offset, fulfillment start offset, and fulfillment completion offset are all returned to zero, the chain tail disturbance is written as a zero state; If any of the following offsets—performance sequence offset, performance start offset, or performance completion offset—is not zeroed, the chain tail disturbance will be written as not zeroed.
[0036] In this embodiment, generating a price reversal proof specifically includes: Read one of the zeroed-out or non-zeroed-out states of the chain tail perturbation; When the chain tail disturbance is in a zero state, the basic acceptance price corresponding to the target order status is written into the acceptance price; In this invention, the basic acceptance price is the initial order acceptance price corresponding to the service type, service location, appointment time, and service duration requirement in the target order status. The basic acceptance price is read from the basic price mapping record saved by the platform and then written into the target order status. The basic price mapping record is a price correspondence record saved by the platform, used to record the correspondence between the service type, service location, appointment time, service duration requirement, and initial order acceptance price. When the chain tail disturbance is not in a zero state, write the basic acceptance price corresponding to the target order status item by item in the order of performance sequence offset, performance start offset, and performance completion offset to obtain the acceptance price; In this invention, writing item by item is a processing operation that writes the offset content in the chain tail disturbance into the basic acceptance price in a fixed order; when the chain tail disturbance is not in a zero state, the performance order offset, performance start offset, and performance completion offset are written into the basic acceptance price in sequence, and the price record after writing is completed is used as the acceptance price. The accepted price is written back to the target order status via the misaligned occupant point, locking in the price acceptance range corresponding to the target order status; In this invention, the acceptance price is written back to the target order status via the misaligned occupant point. This is to write the acceptance price back to the target order status along the chain node sequence number corresponding to the misaligned occupant point, so that the acceptance price is compared with the price acceptance range in the target order status. The price acceptance range is fixed by reading the lower and upper boundaries of the price acceptance range in the target order status after the price is written back. The price acceptance range will not be updated before the current closed comparison is completed. Perform a closed-loop comparison between the accepted price and the price range; In this invention, the closed comparison is used to determine the price range between the acceptance price and the lower boundary and the upper boundary of the price acceptance range. When the price falls into the price acceptance range, write the price reversal to the empty state; When the accepting price is lower than the lower boundary of the price acceptance range, write the low-price overshoot status into the price proof; when the accepting price exceeds the upper boundary of the price acceptance range, write the high-price overshoot status into the price proof.
[0037] In this invention, the set of states for price rebuttal is fixed as an empty state, a low-price out-of-bounds state, and a high-price out-of-bounds state; an empty state indicates that the accepting price meets the price acceptance range; a low-price out-of-bounds state indicates that the accepting price is below the lower boundary of the price acceptance range; and a high-price out-of-bounds state indicates that the accepting price exceeds the upper boundary of the price acceptance range.
[0038] In this embodiment, the freezing and releasing process of the candidate service provider connection edge specifically includes: Read the tail perturbation and price reversal evidence bound to the candidate service provider's connection edge; In this invention, the chain tail disturbance is a verification record formed after the candidate service provider connection edge is sequentially snapped together at the chain tail node, and the price reversal is a verification record formed after the candidate service provider connection edge is closed and compared with the acceptance price and the price acceptance range; reading the chain tail disturbance and price reversal bound to the candidate service provider connection edge means reading the chain tail disturbance state and price reversal state bound to the same candidate service provider connection edge. The chain tail disturbance state is fixed as zero state and non-zero state, and the price reversal state is fixed as empty state, low price out of bounds state and high price out of bounds state. When the chain tail disturbance is not in a zero state, the candidate service provider connection edge is written to the performance freeze state to prevent the candidate service provider connection edge from entering the residual deduction. In this invention, the performance freeze state is the freeze state written by the candidate service provider connection edge due to the chain tail disturbance being in a non-zero state; after the candidate service provider connection edge is written to the performance freeze state, it does not enter the residual decomposition of the synchronous shadow diffusion result and the forward diffusion result. When the price counter-proof is written into either the low price out of bounds state or the high price out of bounds state, the candidate service provider connection edge is written into the pricing freeze state to prevent the candidate service provider connection edge from entering the residual deduction. In this invention, the pricing freeze state is the frozen state written by the candidate service provider connection edge due to the price counter-proof being written into either the low price out-of-bounds state or the high price out-of-bounds state; after the candidate service provider connection edge is written into the pricing freeze state, it does not enter the residual deduction between the synchronous shadow diffusion result and the forward diffusion result. When the chain tail disturbance is in a zero state and the price proof is empty, clear the unverified state of the candidate service provider connection edge and write the candidate service provider connection edge into the connection release state. In this invention, the connection release state is the participating state written when the candidate service provider connection edge is in a zero state at the chain tail perturbation and the price proof is empty; before writing the connection release state, the unverified state of the candidate service provider connection edge is cleared; after writing the connection release state, the candidate service provider connection edge enters the residual decomposition of the synchronous shadow diffusion result and the forward diffusion result. When a candidate service provider's connection edge is written to either the performance freeze state or the pricing freeze state, it remains frozen and connection release is prohibited. In this invention, when the candidate service provider connection edge is already written to either the performance freeze state or the pricing freeze state, the freeze state is not cleared and the connection release state is not written; only the connection release state is allowed to enter the residual deduction, so that the candidate service provider connection edge state is read before the residual deduction, and the candidate service provider connection edge that is not in the connection release state is not sent to the residual deduction. Only candidate service provider connection edges in the connection release state are allowed to enter the residual subtraction of synchronous shadow diffusion results and forward diffusion results.
[0039] In this implementation, the service provider allocation result and order price are generated based on the residual deduction result, specifically including: Using candidate service provider connection edges in the connection release state as the snapping objects, extract the synchronous shadow diffusion results and forward diffusion results bound to the same candidate service provider connection edges; Press the synchronous shadow diffusion results and forward diffusion results into the same sequence slot according to the same chain node number, and prohibit the out-of-order snapping between different chain node numbers; In this invention, the synchronous locking slot is the locking position where the synchronous shadow diffusion result and the forward diffusion result are written according to the same chain node number; the synchronous locking slot is set according to the chain node number, and one chain node number corresponds to one synchronous locking slot; the synchronous shadow diffusion result and the forward diffusion result with different chain node numbers must not be written into the same synchronous locking slot. Out-of-order snapping refers to snapping operations that occur between different chain node numbers; in this invention, prohibiting out-of-order snapping means that only synchronous shadow diffusion results and forward diffusion results under the same chain node number are allowed to enter the same out-of-order snapping slot. Using the vacancy mark corresponding to the chain node number as the residual entry, the vacancy propagation state of the misplaced vacancy point in the forward diffusion result is matched with the shadow propagation state of the chain nodes before and after the vacancy mark in the synchronous shadow diffusion result. In this invention, the residual entry is the chain node number corresponding to the empty position mark; the residual entry is used to determine the snapping start point before and after the target order is inserted, so that the positive diffusion result corresponding to the misaligned occupancy point and the synchronous shadow diffusion result corresponding to the chain node before and after the empty position mark snap together under the same chain node number. The placeholder propagation state is the state record of the chain node corresponding to the misplaced placeholder in the forward diffusion result, which comes from the latch result after the placeholder chain mirror performs forward diffusion; the shadow propagation state is the state record of the chain node corresponding to the empty space marker before and after in the synchronous shadow diffusion result, which comes from the latch result after the original chain mirror performs synchronous shadow diffusion. Write the propagation offset that is not fully locked after locking into the residual locking result. When the propagation offset returns to zero, write the residual locking result into the residual closed state. In this invention, the propagation offset is the difference between the occupant propagation state and the shadow propagation state in terms of arrival order, start time, and completion time, including arrival order offset, start time offset, and completion time offset; when all three offsets are zero, the propagation offset is zero; if any one of them is not zero, the propagation offset is not zero. The residual closure state is the closure state written when the propagation offset in the residual snapping result returns to zero; after the candidate service provider connection edge writes the residual closure state, it means that the insertion of the target order into the candidate service chain has not disrupted the snapping relationship between the synchronous shadow diffusion result and the forward diffusion result at the residual entry point; Perform edge selection processing on the candidate service provider connection edges that are written to the residual closure state, and write the candidate service providers corresponding to the selected candidate service provider connection edges into the service provider allocation results; In this invention, the service provider allocation result is the candidate service provider record determined based on the residual deduction result; The acceptance price corresponding to the selected candidate service provider's connection edge is written into the order price.
[0040] In this invention, the order price is the acceptance price bound to the candidate service provider connection edge corresponding to the residual closure state. During generation, the candidate service providers bound to the candidate service provider connection edge corresponding to the residual closure state are written into the service provider allocation result, and the acceptance price bound to the candidate service provider connection edge corresponding to the residual closure state is written into the order price.
[0041] In this embodiment, the edge selection process specifically includes: Read and write candidate service provider connection edges in the residual closure state; Candidate service provider connection edges are written into the closed sequence according to the order in which the residual closure state is written, and candidate service provider connection edges that have not been written into the residual closure state are prohibited from entering the closed sequence. In this invention, the closed sequence is a queue formed by arranging the candidate service provider connection edges that have been written with residual closed states in the order in which they are written with residual closed states; the candidate service provider connection edges that have been written with residual closed states earlier are placed at the beginning of the closed sequence, and the candidate service provider connection edges that have not been written with residual closed states are not written with the closed sequence. When there are several candidate service provider connection edges in the same writing order, the candidate service provider connection edge whose residual entry is written first is selected according to the writing order of the residual entry in the same sequence locking slot. In this invention, the writing order is the batch order formed when the residual closed state is written into the closed sequence; several candidate service provider connection edges that write the residual closed state in the same residual closed state writing operation are in the same writing order; when several candidate service provider connection edges are in the same writing order, the writing order of the residual entrance in the same sequence locking slot is read, and the candidate service provider connection edge that writes the residual entrance first is selected first. The selected candidate service providers will be locked as the target connection edges; In this invention, the target connection edge is the candidate service provider connection edge locked after being determined by the closed sequence and the order of writing the residual entry; the target connection edge is used to determine the service provider allocation result and the order price; the order of writing the residual entry is the order in which the empty space marker corresponding to the chain node number is written into the same-order locking slot when the candidate service provider connection edge enters the same-order locking slot. Write the candidate service providers corresponding to the target connection edge into the service provider allocation results; Write the acceptance price corresponding to the target connection edge into the order price.
[0042] Example 1: To verify the feasibility of this invention in practice, it was applied to the daily scheduling and pricing scenario of a comprehensive on-site service platform. The service orders accepted by the platform include on-site cleaning, equipment repair, installation and debugging, and maintenance. After a user submits a service request on the platform, the platform generates a target order status, which records the service type, service location, appointment time, service duration requirement, and acceptable price range. The platform already has several candidate service providers, each with accepted but not yet completed tasks. The tasks are ordered according to the confirmed appointment time, the platform's dispatch time, and the time they are added to the candidate service chain. Traditional rule-based scheduling mainly dispatches tasks based on distance, availability, and service type matching, easily overlooking the impact of new orders on the tail nodes of subsequent task chains for candidate service providers. In actual operation, common problems fall into three categories: new orders cause subsequent order start times to be delayed; the platform's offered price exceeds the user's acceptable range; and candidate service providers are reassigned due to fulfillment conflicts, resulting in longer waiting times and a decreased user experience.
[0043] In this embodiment, after receiving a target order, the platform does not directly determine the service provider based on distance or availability. Instead, it first reads the target order status and the candidate service chain. The system uses the target order status as a node to be inserted, and writes the tasks accepted but not completed by the candidate service providers as chain nodes. Adjacent chain nodes are connected according to the order of fulfillment, constructing a dynamic supply and demand graph containing connection edges of candidate service providers. The platform writes a pending verification status to each candidate service provider connection edge and searches for insertable chain segments along the candidate service chain. If the service completion time of the front-end chain node is no later than the target order reservation time, and the target order reservation time plus service duration requirement is no later than the service start time of the back-end chain node, then the adjacent chain nodes are marked as insertable chain segments. The node to be inserted is then attached to the front-end and back-end chain nodes, forming a pending verification channel.
[0044] Once the verification channel is established, the target order is pushed between the front-end and back-end chain nodes and marked as a misalignment placeholder. The platform retains the original fulfillment order connection between the front-end and back-end chain nodes and writes the front-end placeholder order connection and the back-end placeholder order connection. The service chain before embedding is copied as a mirror image of the original chain, and the service chain after embedding is copied as a mirror image of the placeholder chain. The original chain mirror is used to record the fulfillment propagation reference when the target order is not inserted, and the placeholder chain mirror is used to record the fulfillment propagation path after the target order is inserted. The two mirror chains use the same chain node sequence number, and the position in the original chain mirror corresponding to the misalignment placeholder is written with an empty slot mark to form a same-order comparison structure.
[0045] After the same-order comparison structure enters the DCRNN diffusing convolutional recurrent unit, the platform locks the same diffusing round, diffusing stride, diffusing order, and diffusing weight for the original chain mirror and the placeholder chain mirror. The original chain mirror performs synchronous shadow diffusing along the fulfillment order connection, while the placeholder chain mirror performs forward diffusing along the fulfillment order connection, the first part of the placeholder order connection, the misaligned placeholder point, and the last part of the placeholder order connection. At the end of each diffusing round, the system synchronously latches the diffusing states of the original chain mirror and the placeholder chain mirror according to the same chain node number, obtaining the synchronous shadow diffusing result and the forward diffusing result. After the forward diffusing result hits the misaligned placeholder point for the first time, the platform uses the tail node of the placeholder chain mirror as the reverse diffusing entry point, and reverses back to the misaligned placeholder point along the last part of the placeholder order connection and the fulfillment order connection. At the tail node, the forward diffusing result is synchronously merged with the synchronous shadow diffusing result under the same chain node number.
[0046] Sequential offsetting forms the performance sequence offset, performance start offset, and performance completion offset. When all three types of offsets are zero, the chain tail perturbation is written to a zero state; if any one is not zero, the chain tail perturbation is written to a non-zero state. Candidate service provider connection edges with a non-zero chain tail perturbation are written to the performance freeze state and no longer enter residual offsetting. When the chain tail perturbation is zero, the platform reads the basic acceptance price corresponding to the target order state; when the chain tail perturbation is not zero, the platform writes the performance sequence offset, performance start offset, and performance completion offset to the basic acceptance price in sequence to form the acceptance price. After the acceptance price is written back to the target order state through the misaligned occupant point, a closed comparison is performed with the price acceptance range. When the acceptance price falls into the price acceptance range, the price rebuttal is written to an empty state; when the acceptance price is lower than the lower boundary, the price rebuttal is written to a low price out of bounds state; when the acceptance price exceeds the upper boundary, the price rebuttal is written to a high price out of bounds state. Candidate service provider connection edges with price rebuttal are written to a pricing freeze state.
[0047] After fulfillment and pricing freezes, only candidate service provider connection edges with zero tail disturbance and an empty price are written into the connection release state. The platform performs residual subtraction between the synchronous shadow diffusion result and the forward diffusion result for the candidate service provider connection edges in the connection release state. During the residual subtraction process, the system uses the chain node number corresponding to the empty position mark as the residual entry, and subtracts the occupancy propagation state corresponding to the misaligned occupancy point to the shadow propagation state of the chain nodes before and after the empty position mark. When the propagation offset reaches zero, the residual subtraction result is written into the residual closure state. Candidate service provider connection edges written into the residual closure state enter the closure sequence; if several candidate service provider connection edges are in the same writing order, the candidate service provider connection edge whose residual entry is written first is selected according to the writing order of the residual entry in the same order subtraction slot, and locked as the target connection edge. The platform writes the candidate service provider corresponding to the target connection edge into the service provider allocation result, and writes the acceptance price corresponding to the target connection edge into the order price.
[0048] In a sample of similar business volume, 1200 orders were processed using conventional rules and 1200 orders were processed using this invention. Conventional rule processing primarily relied on distance, service type matching, and idle status. This invention added features such as misalignment, mirror diffusion, chain tail disturbance, price reversal, freeze release, residual deduction, and edge selection before order dispatch. Comparative data shows that this invention reduces the impact of order insertion on chain tail nodes, lowers the re-dispatch rate, and improves price acceptance and on-time fulfillment rates. Especially in scenarios where candidate service providers already have continuous task schedules, this invention can preemptively exclude candidate connections with non-zeroed fulfillment order offsets, start time offsets, and completion time offsets before service provider allocation, preventing service providers that appear capable of accepting orders but have excessive subsequent fulfillment pressure from being included in the allocation results.
[0049] Table 1: Comparison of Scheduling and Pricing Effects Before and After Implementation of the Invention
[0050] As shown in Table 1, under the condition of maintaining a consistent sample order volume, this invention significantly improves order dispatch speed, fulfillment stability, and price acceptance. The data in the table is derived from a comparison of continuous operation records under samples of similar business volume. The results of conventional rule processing serve as a control. The processing results of this invention are obtained using dynamic supply and demand graphs, misalignment, mirror diffusion, chain tail disturbance verification, price reversal verification, and residual deduction edge selection. The average order dispatch time decreased from 42.6 seconds to 18.9 seconds, indicating that this invention did not slow down the scheduling process due to the introduction of mirror diffusion and residual deduction. This is because candidate service provider connection edges have already been screened through fulfillment freezing and pricing freezing before entering residual deduction. Subsequent processing only involves candidate service provider connection edges in the connection release state, reducing the number of invalid candidate connections participating in the final allocation.
[0051] The early interception rate of fulfillment conflicts increased from 58.4% to 91.7%, which directly reflects the effectiveness of the tail-end disturbance mechanism. Conventional rule processing often only considers whether the service provider is currently idle, making it difficult to detect the impact of inserting a target order on the tail nodes of subsequent tasks. This invention, by synchronously combining the shadow diffusion results and the forward diffusion results, writes the fulfillment order offset, fulfillment start offset, and fulfillment completion offset into the tail-end disturbance. When the tail-end disturbance is not zeroed out, it directly writes the fulfillment freeze state, thus intercepting candidate service provider connection edges that are not suitable for order insertion before allocation.
[0052] Price-related data also supports the pricing effectiveness of this invention. The price acceptance rate increased from 68.4% to 82.7%, and the accuracy rate of price overrun interception increased from 72.1% to 92.4%. This indicates that after the accepted price is written back to the target order status through the misaligned occupant point, it is compared with the closed price acceptance range, which can identify low-price and high-price overruns earlier. Under conventional rules, some orders only reveal price mismatches after dispatch, which easily leads to cancellations or reassignments. This invention writes price counter-evidence into the candidate service provider connection edge status. When price counter-evidence exists, the pricing is directly frozen, reducing the probability of price-mismatched connections entering subsequent edge selection processing.
[0053] The reassignment rate decreased from 9.8% to 3.2%, while the on-time fulfillment rate increased from 84.1% to 93.8%, indicating that this invention not only improves the accuracy of service provider selection but also reduces the impact of order insertion on later tasks in the service chain. The residual closure state requires the propagation offset to reach zero before allowing edge selection processing, incorporating the fulfillment chain stability and acceptance price of candidate service providers into the allocation judgment. The above data shows that this invention solves the problem of "seemingly acceptable orders that actually affect subsequent fulfillment" in conventional rule-based order assignment, making the service provider allocation results and order prices more closely reflect the actual fulfillment status.
[0054] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A dynamic scheduling and pricing method for on-site services based on multidimensional data and machine learning, characterized in that, Includes the following steps: Obtain the target order status and candidate service chain, and construct a dynamic supply and demand graph containing candidate service provider connection edges according to the on-site service fulfillment order; The target order is embedded into the candidate service chain as a misaligned placeholder, and the service chain before embedding is mirrored as the original chain mirror. The original chain image and the placeholder chain image are fed into the DCRNN diffusing convolution recurrent unit. The original chain image is driven to perform synchronous shadow diffusion with the same diffusion parameters, and the placeholder chain image is driven to perform forward diffusion, so as to obtain the synchronous shadow diffusion result and the forward diffusion result. After forward diffusion reaches the misaligned occupant point, it triggers tail-end reverse diffusion, which returns from the tail end of the occupant chain mirror to the misaligned occupant point, and reads the chain tail perturbation formed by the occupant chain mirror relative to the original chain mirror at the chain tail node. After the chain tail disturbance is folded in, the acceptance price is generated. The acceptance price is then written back to the target order status through the misaligned occupancy point. When the acceptance price does not fall into the price acceptance range corresponding to the target order status, a price counter-proof is generated. For candidate service provider connection edges where the chain tail disturbance has not returned to zero, the performance freeze will be executed; for candidate service provider connection edges where there is a price counter-proof, the pricing freeze will be executed; for candidate service provider connection edges where the chain tail disturbance has returned to zero and there is no price counter-proof, the connection will be released. After the connection is released, perform residual deduction between the synchronous shadow diffusion result and the forward diffusion result on the candidate service provider connection edge, and generate the service provider allocation result and order price according to the residual deduction result.
2. The method for dynamic scheduling and pricing of on-site services based on multidimensional data and machine learning according to claim 1, characterized in that, The process of constructing the dynamic supply and demand map specifically includes: Using the target order status as the node to be inserted, candidate service chains are developed according to the order of execution of unfulfilled tasks; Write the unfinished tasks in the candidate service chain as chain nodes in sequence, and connect adjacent chain nodes according to the order of execution. Mark insertable chain segments between adjacent chain nodes along the candidate service chain, and preserve the order of performance between adjacent chain nodes; The node to be inserted is attached to the front and back chain nodes of the insertable chain segment, respectively, so that the insertable chain segment forms a channel to be verified. Bind the channel to be verified to the candidate service provider's connection edge, and write the verification status to the candidate service provider's connection edge; Candidate service chains that have not yet formed a verification channel are prohibited from entering the mirroring process of the original chain mirror and the placeholder chain mirror.
3. The method for dynamic scheduling and pricing of on-site services based on multidimensional data and machine learning according to claim 2, characterized in that, The step of embedding the target order as a misaligned occupant point into the candidate service chain specifically includes: Read the candidate service provider connection edge that is in the pending verification state and locate the pending verification channel bound to the candidate service provider connection edge; Read the front-end chain node and the back-end chain node along the channel to be verified, and push the node to be inserted between the front-end chain node and the back-end chain node; Mark the inserted node as a misalignment occupant point and maintain the order of fulfillment from the front-end chain node to the back-end chain node without deleting it; Write a front-end placeholder sequential connection between the front-end chain node and the misaligned placeholder point, and write a back-end placeholder sequential connection between the misaligned placeholder point and the back-end chain node. Process the unverified channels in the candidate service chain one by one according to the order of performance, so that each unverified channel forms a placeholder candidate chain, and bind the placeholder candidate chain to the corresponding candidate service provider connection edge.
4. The method for dynamic scheduling and pricing of on-site services based on multidimensional data and machine learning according to claim 3, characterized in that, The process of mirroring the pre-embedded service chain into the original chain image and the post-embedded service chain into a placeholder chain image specifically includes: Read the candidate service chain and placeholder candidate chain bound to the candidate service provider connection edge; Copy all chain nodes and fulfillment order connections in the candidate service chain to obtain a mirror image of the original chain; Copy all chain nodes, misaligned placeholder points, fulfillment order connections, front-end placeholder order connections, and back-end placeholder order connections in the placeholder candidate chain to obtain a placeholder chain mirror image. Bind the original chain image and the placeholder chain image to the same candidate service provider connection edge; Write the same chain node sequence number to the original chain mirror and the placeholder chain mirror according to the order of performance; Write an empty mark at the corresponding misaligned position of the chain node in the original chain mirror, and retain the order of the chain nodes before and after the empty mark so that the original chain mirror and the occupied chain mirror form a same-order comparison structure. The same-order comparison structure is bound to the candidate service provider connection edge and written into the same-parameter input queue of the DCRNN diffusing convolution recurrent unit.
5. The method for dynamic scheduling and pricing of on-site services based on multidimensional data and machine learning according to claim 4, characterized in that, The obtained synchronous shadow diffusion results and forward diffusion results specifically include: Read the same-order comparison structure of the connection edges bound to the same candidate service provider in the same parameter input queue; Lock the diffusion rounds, diffusion step size, diffusion order, and diffusion weight of the original chain mirror and the placeholder chain mirror; The original chain mirror is driven to perform synchronous shadow diffusion along the fulfillment order connection. When diffusion reaches the empty slot mark, the fulfillment order connection of the chain nodes before and after the empty slot mark is maintained. The drive placeholder chain mirrors are connected along the fulfillment sequence, the front placeholder sequence is connected, the staggered placeholder points are connected, and the rear placeholder sequence is connected to perform forward diffusion; At the end of each diffusion round, the original chain mirror diffusion state and the placeholder chain mirror diffusion state under the same chain node number are synchronously latched. Diffusion states with different chain node numbers are prohibited from being written to the same latch location. Write the original chain mirror diffusion state after synchronization latch as the synchronization shadow diffusion result, and write the placeholder chain mirror diffusion state after synchronization latch as the forward diffusion result.
6. The method for dynamic scheduling and pricing of on-site services based on multidimensional data and machine learning according to claim 5, characterized in that, The generation process of the chain tail perturbation specifically includes: Capture the diffusion round in which the forward diffusion result first hits the misaligned site, and lock that diffusion round as the tail-end reverse diffusion trigger round; At the tail end, the reverse diffusion triggers a round of latching the arrival position corresponding to the forward diffusion result, and prohibits the forward diffusion result after the misaligned occupancy point from being directly written into the chain tail perturbation; Using the tail node of the placeholder chain mirror as the reverse diffusion entry point, connect along the fulfillment sequence of the placeholder chain mirror and then connect the subsequent placeholder sequence back to the misaligned placeholder point. At the tail node, the forward diffusion result is sequentially merged with the synchronous shadow diffusion result under the same chain node number; Write the performance sequence offset, performance start offset, and performance completion offset formed by the same sequence of snapping into the chain tail disturbance; When the fulfillment sequence offset, fulfillment start offset, and fulfillment completion offset are all returned to zero, the chain tail disturbance is written as a zero state; If any of the following offsets—performance sequence offset, performance start offset, or performance completion offset—is not zeroed, the chain tail disturbance will be written as not zeroed.
7. The method for dynamic scheduling and pricing of on-site services based on multidimensional data and machine learning according to claim 6, characterized in that, The generation of price reversal proof specifically includes: Read one of the zeroed-out or non-zeroed-out states of the chain tail perturbation; When the chain tail disturbance is in a zero state, the basic acceptance price corresponding to the target order status is written into the acceptance price; When the chain tail disturbance is not in a zero state, write the basic acceptance price corresponding to the target order status item by item in the order of performance sequence offset, performance start offset, and performance completion offset to obtain the acceptance price; The accepted price is written back to the target order status via the misaligned occupant point, locking in the price acceptance range corresponding to the target order status; Perform a closed-loop comparison between the accepted price and the price range; When the price falls into the price acceptance range, write the price reversal to the empty state; When the accepting price is lower than the lower boundary of the price acceptance range, write the low-price overshoot status into the price proof; when the accepting price exceeds the upper boundary of the price acceptance range, write the high-price overshoot status into the price proof.
8. The method for dynamic scheduling and pricing of on-site services based on multidimensional data and machine learning according to claim 7, characterized in that, The freeze-release process for the candidate service provider connection edge specifically includes: Read the tail perturbation and price reversal evidence bound to the candidate service provider's connection edge; When the chain tail disturbance is not in a zero state, the candidate service provider connection edge is written to the performance freeze state to prevent the candidate service provider connection edge from entering the residual deduction. When the price counter-proof is written into either the low price out of bounds state or the high price out of bounds state, the candidate service provider connection edge is written into the pricing freeze state to prevent the candidate service provider connection edge from entering the residual deduction. When the chain tail disturbance is in a zero state and the price proof is empty, clear the unverified state of the candidate service provider connection edge and write the candidate service provider connection edge into the connection release state. When a candidate service provider's connection edge is written to either the performance freeze state or the pricing freeze state, it remains frozen and connection release is prohibited. Only candidate service provider connection edges in the connection release state are allowed to enter the residual subtraction of synchronous shadow diffusion results and forward diffusion results.
9. The method for dynamic scheduling and pricing of on-site services based on multidimensional data and machine learning according to claim 8, characterized in that, The process of generating service provider allocation results and order prices based on residual deduction results specifically includes: Using candidate service provider connection edges in the connection release state as the snapping objects, extract the synchronous shadow diffusion results and forward diffusion results bound to the same candidate service provider connection edges; Press the synchronous shadow diffusion results and forward diffusion results into the same sequence slot according to the same chain node number, and prohibit the out-of-order snapping between different chain node numbers; Using the vacancy mark corresponding to the chain node number as the residual entry, the vacancy propagation state of the misplaced vacancy point in the forward diffusion result is matched with the shadow propagation state of the chain nodes before and after the vacancy mark in the synchronous shadow diffusion result. Write the propagation offset that is not fully locked after locking into the residual locking result. When the propagation offset returns to zero, write the residual locking result into the residual closed state. Perform edge selection processing on the candidate service provider connection edges that are written to the residual closure state, and write the candidate service providers corresponding to the selected candidate service provider connection edges into the service provider allocation results; The acceptance price corresponding to the selected candidate service provider's connection edge is written into the order price.
10. The method for dynamic scheduling and pricing of on-site services based on multidimensional data and machine learning according to claim 9, characterized in that, The edge selection process specifically includes: Read and write candidate service provider connection edges in the residual closure state; Candidate service provider connection edges are written into the closed sequence according to the order in which the residual closure state is written, and candidate service provider connection edges that have not been written into the residual closure state are prohibited from entering the closed sequence. When there are several candidate service provider connection edges in the same writing order, the candidate service provider connection edge whose residual entry is written first is selected according to the writing order of the residual entry in the same sequence locking slot. The selected candidate service providers will be locked as the target connection edges; Write the candidate service providers corresponding to the target connection edge into the service provider allocation results; Write the acceptance price corresponding to the target connection edge into the order price.
